Scope · Official topics → our modules

Course Syllabus

Every topic on the official USAAIO syllabus, organized into 8 teachable modules. Round 1 covers Modules 0–5; Round 2 covers everything. 官方考纲拆分为 8 个模块。第一轮覆盖 0–5，第二轮覆盖全部。

The core principle USAAIO is neither pure math nor pure coding. For every topic Harper must know both the theory (derive it by hand) and the programming (build it from scratch in NumPy/PyTorch). Black-box use of scikit-learn is not enough.

Important dates · 2027 cycle 重要日期

Date	Event	Details
Jun 1, 2026	Registration opens 报名开始	Register on the usaaio.org portal, then find a proctor. Round 1 is open to everyone.
Jan 31, 2027 11:59 p.m. EST	Registration closes 报名截止	Must be registered and have a proctor arranged before Round 1.
Fri, Feb 19, 2027 12:00–3:00 p.m. ET	Round 1 第一轮	At school or a USAAIO-authorized test site. Part 1 · Non-coding — start 12:00–12:15, 60 min, multiple-choice / fill-in-the-blank; closes 1:15 p.m. sharp. Part 2 · Coding — start 1:30–1:45, 75 min; closes 3:00 p.m. sharp. 12:00 p.m. ET = 1:00 a.m. Sat, Feb 20 in Taipei / Beijing.
Sat, Feb 20, 2027 11:59 p.m. ET	Supplementary documents due 补充材料截止	Submit URLs for all required documents, including the OBS screen recording.
Mar – Apr 2027	Round 2 第二轮	By qualification through Round 1. Full syllabus; some problems use GPUs. Exact date and site not yet announced (2026 Round 2 was held at Harvard & MIT).
Jun 2027	USAAIO Camp 集训营	Top Round 2 performers; Team USA for IOAI selected from camp.

Source: usaaio.org/2027-usa-na-aio (checked Sep 28, 2026). Round 2 and camp dates will be updated when announced.

Official USA-NA-AIO syllabus (reference) 官方考纲原文

Reproduced from the official syllabus as a reference. Our module grouping follows below.

Markdown programming in Google Colab

Some questions require contestants to write their solutions in Google Colab text cells using Markdown (for example, to enter mathematical equations). Contestants need to know how to enter text, write code snippets, and typeset mathematical formulas.

Module 0

Mathematical foundations for AI

Linear algebra (e.g., affine transformations, matrix decompositions, eigenvalues and eigenvectors)
Probability and statistics (e.g., Bayes' rule and Hoeffding's inequality)
Derivatives in multivariable calculus
Convex optimization (e.g., gradient descent algorithms and duality)

Module 1

Basic coding

Python
NumPy
pandas
matplotlib.pyplot
seaborn
scikit-learn

Module 0 · 2

Machine learning

Supervised learning (e.g., linear regression, logistic regression, support vector machine, decision trees, kNN, ensemble learning, bias-variance tradeoff, cross-validation, loss functions)
Unsupervised learning (e.g., k-means clustering, principal component analysis)

Module 2 · 3

Advanced coding for deep learning (PyTorch)

In USA-NA-AIO, deep learning problems must be programmed with PyTorch rather than TensorFlow. This is consistent with IOAI requirements and current trends in academia and industry.

Module 4

Deep learning foundations

Multilayer perceptron models
Essential layers (e.g., affine transformations, batch normalization, and dropout)
Forward propagation, backpropagation, and their mathematical computations (by hand)

Module 4

Transformers

Note: Transformers are the foundation of many modern AI technologies. Therefore, contestants must have a thorough understanding of transformers.

Attention mechanisms
Transformer architecture
Applications (e.g., natural language processing, vision transformers, and graph neural networks)

Module 6

Natural language processing

Tokenization
Word embeddings
Transformers
Pre-training
Fine-tuning

Module 6

Computer vision and generative AI

Convolutional neural network
Object detection
UNet
Autoencoder
Variational autoencoder
Generative adversarial network
Denoising diffusion probabilistic models
Stable diffusion

Module 5 · 7

Module map · official topics → our modules

#	Module	Round	Official topics covered
0	Python & Data Tooling	R1	Python, NumPy, pandas, matplotlib.pyplot, seaborn; Markdown in Google Colab
1	Math Foundations for AI	R1	Linear algebra (affine transforms, matrix decompositions, eigenvalues/eigenvectors); probability & statistics (Bayes' rule, Hoeffding's inequality); multivariable derivatives; convex optimization (gradient descent, duality)
2	Supervised Learning	R1	Linear & logistic regression, SVM, decision trees, kNN, ensemble learning, bias-variance tradeoff, cross-validation, loss functions
3	Unsupervised Learning	R1	k-means clustering, principal component analysis (PCA)
4	Deep Learning Foundations	R1	Multi-layer perceptron; essential layers (affine, batch norm, dropout); forward & backpropagation by hand; PyTorch
5	Convolutional Neural Networks	R1	CNN basics, image tasks (Round 1 intro level)
6	Transformers & NLP	R2	Attention, transformer architecture, vision transformers, GNNs; tokenization, word embeddings, pre-training, fine-tuning
7	Computer Vision & Generative AI	R2	Object detection, UNet, autoencoder, VAE, GAN, denoising diffusion (DDPM), stable diffusion

Round breakdown

Topic coverage per round as stated by USAAIO at usaaio.org/2027-usa-na-aio#round-1.

Round 1 — Fri, Feb 19, 2027

Modules 0–5

Official topics: Markdown programming in Google Colab; Mathematical foundations for AI; Basic coding; Machine learning; Advanced coding for deep learning (PyTorch); Deep learning foundations; Basics of convolutional neural networks (CNNs).

Format: Google Colab, multiple multi-part problems. Some parts are non-coding (typeset math/reasoning in text cells with Markdown); some are coding (code cells). “All code must run on CPUs. In Round 1, GPUs are neither required nor allowed.” 3 hours (Part 1 non-coding 60 min + Part 2 coding 75 min), proctored at a school or authorized test site.

Round 2 — Mar–Apr 2027

Everything (Modules 0–7)

Official topics: “All syllabus topics”, adding Transformers, NLP, and the rest of computer vision & generative AI.

Same format as Round 1, except some problems may require GPUs (Colab L4). Qualify via Round 1. Transformers are flagged as needing especially deep understanding.

What "knowing a topic" means here

Theory side

e.g. derive the least-squares estimator in linear regression; prove a matrix is a valid (positive-definite) kernel; compute backprop gradients by hand.

Programming side

e.g. build a PCA class from scratch in NumPy via the eigenvalue problem; build a fully-connected network from scratch and explain every step; implement training loops in PyTorch.

Math Academy fits here Math Academy covers Module 1's prerequisites efficiently (linear algebra, probability, multivariable calculus). See the Math Track page for the exact course mapping.

Sources: official USAAIO syllabus (usaaio.org/syllabus); round coverage from usaaio.org/2027-usa-na-aio#round-1. Module grouping is this course's design.